Graduate, current student present SOU research at AI conference

(Ashland, Ore.) — Southern Oregon University computer science graduate Chandler Campbell and current student Jacob Nowack attended a conference of research software engineers in Philadelphia last month to showcase their work on a pair of closely related projects that hinge on the use of artificial intelligence to simplify and organize highly complex research tasks.

Campbell presented a paper on his study of tacit knowledge in research settings – gathering, storing and retrieving the unspoken practices of academic teams that sometimes are lost when a project is disrupted or ends – and Nowack spoke about using a tacit knowledge tool to help UCLA astronomers rapidly expand their efforts to survey billions of distant galaxies. Both Campbell and Nowack work on their AI projects under Bernadette Boscoe, an SOU assistant professor of computer science who builds and researches infrastructures and tools to help domain scientists do their work.

Campbell and Nowack were presenters at the third annual national conference of the US Research Software Engineer Association, an organization that supports those who use expertise in programing to advance research. The association is a project of a California-based nonprofit.

“I got to meet a lot of really interesting people from all over the country, and learned more about cutting-edge AI technologies and software development techniques which I think will help me a lot in my future career,” Nowack said, describing his experience at the conference.

“I was a bit nervous initially going into it, but when the time came I had a great time presenting,” he said.

Nowack’s project is intended to help astronomers who measure the distances to far-flung galaxies so they can better understand how the universe has expanded and evolved. Spectroscopy, the traditional method of measuring those distances, is expensive and time-consuming.

“Our project uses machine learning to solve this problem,” Nowack said. “We trained an AI model on approximately 286,000 galaxies whose distances were already measured using spectroscopy. Once trained, (the AI model) can estimate distances over 1,000 times faster than traditional spectroscopy, making large-scale cosmic surveys practical.”

His work with the UCLA astronomers is based on a Large Language Model (LLM) of artificial intelligence that is used to archive the group’s protocols.

Boscoe’s research group at SOU has developed a Retrieval-Augmented Generation (RAG) system – an AI framework that pairs an LLM with an information retrieval system to improve accuracy and relevance of resulting data. Her research has received grants over the past two years from the Sloan Foundation and the National Science Foundation, and Boscoe has worked with Campbell to build the project’s RAG-LLM tool – AquiLLM, which was the subject of Campbell’s presentation at last month’s conference.

“Our work on AquiLLM is part software development and part social research,” he said. “We’re investigating the potential for an AI-enabled knowledge repository to improve how academic research groups function.”

Tacit knowledge – which can include informal practices such as notes, meeting transcripts and group communications – can sometimes be lost when participants come and go from academic research groups.

“Our hope is that if we can ingest enough informal communication into the system, and give an LLM access to that information, it will be able to help group members access the tacit knowledge of the group,” Campbell said. “To do this, we’ve written a custom Retrieval-Augmented Generation tool (AquiLLM) specifically for researchers. We have a beta version deployed for astronomers at UCLA, and we’re currently working on fleshing out more functionality.”

The Philadelphia conference included representatives from several national research labs and dozens of top research universities, and Campbell said many were thinking about the same issues that his work addresses.

“I got a lot of valuable feedback on our work, and got to see how other researchers are trying to use AI to solve adjacent problems,” he said. “I was very proud to be there in the mix, representing SOU.”

-SOU-